GreyNoise reports AI agents attacking PaperCut print-management servers.
Source: GreyNoiseGreyNoise disclosed this campaign on September 9. PaperCut's August 27 warning predates it; the earlier customer incident is not established as the same campaign.
Editorial analogy: a print server is a service room inside an office. Breaking into that room can expose keys to other rooms if its permissions are too broad.
An analogy for the mechanism; not an additional claim about the incident.At least 440 installations were compromised; 12 organizations saw attackers gain control over Windows network accounts. GreyNoise
Codex ran a DeepSeek model; this was not OpenAI model use.
Intrusion counts do not mean complete organizational takeover.
Follow-on ransomware or extortion was not established.
PaperCut did not independently verify GreyNoise's indicators.
Editorial lesson: the time between a public fix and an attempted break-in can matter as much as the weakness itself. Restricting what a routine server can access can help contain the consequences.
PaperCut issued security fixes and containment guidance. PaperCut
Editorial defense mapping: fix exposed software and contain the reach of an intrusion.
Keep an inventory of externally reachable applications, apply verified security fixes, and check that the corrected version is running. Give routine servers only the permissions they need.
Installing a fix does not remove access an attacker already obtained. This report does not establish that every affected organization lacked these controls.
Connect unusual server activity with changes to accounts and access rights. Give responders a tested way to isolate a server, review affected credentials and investigate possible wider access.
Reliable records and investigation are needed to establish the extent of a compromise. A breached server does not establish how much of an organization was reachable.
Editorial connections to relevant controls, not evidence that a particular technology would have prevented this case. Each guide links to the security guidance behind its recommendations.
Explore the full AI security frameworkExplore more original accounts in the source report library ↗.
These are source-reported findings. An independent assessment, when available, is labeled explicitly.
01Read this case’s update history →
Reviewed Sep 13, 2026 · Editorial methodology · Structured data